Research map / Image generation / Image restoration
One Network to Solve Them All -- Solving Linear Inverse Problems Using Deep Projection Models
J. H. Rick Chang, Chun-Liang Li, Barnabas Poczos, B. V. K. Vijaya Kumar, Aswin C. Sankaranarayanan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use smooth ELU activations and truncated weights in the classifiers instead of non-smooth leaky ReLU (ablation) | deep projection model within ADMM · Scattered inpainting on ImageNet | RMSE of x - z (ADMM convergence) | Improved |
| replace the wavelet l1 prior with the learned projection network | ADMM solver · Compressive sensing at m/d 0.1 and 0.03 | reconstruction quality | Improved |
| replace the wavelet l1 prior with the learned projection network | ADMM solver · Blockwise inpainting on MS-Celeb-1M | reconstruction quality | Improved |
| use the single projection network instead of a specially-trained network when the measurement matrix is partially resampled | compressive sensing · Resampled Gaussian matrix A on MS-Celeb-1M | reconstruction quality | Improved |
| use the single projection network instead of a specially-trained network under measurement noise | compressive sensing · Gaussian noise on measurements | reconstruction quality | Improved |
| use the single projection network versus BM3D for denoising | ImageNet-trained projector · Image denoising with phone-camera test images | denoising quality: outperforms BM3D when sigma is larger than 40 | Mixed |
| add the projection cost term (v - P(v)) rather than only reconstruction loss (ablation) | projection network training · ADMM iterations | ADMM stability | Improved |
| add spatially varying noise and smoothing perturbations instead of plain Gaussian noise (ablation) | projection network training · ADMM inputs | generalization of projector | Improved |
| increase rho (ablation) | ADMM with projector · Blockwise inpainting on ImageNet | convergence and output quality | Mixed |
| add the latent-space classifier D_l (ablation) | projection network training · projector training | training stability and overfitting | Improved |
Closest papers by meaning
- Untrained Neural Network Priors for Inverse Imaging Problems: A Survey (TPAMI 2022 · similarity 0.88)
- When to Use Convolutional Neural Networks for Inverse Problems (CVPR 2020 · similarity 0.88)
- Deep Feedback Inverse Problem Solver (ECCV 2020 · similarity 0.88)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.86)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.86)
- What's in a Prior? Learned Proximal Networks for Inverse Problems (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
